Particle filter (PF) is widely used in nonlinear/non-Gaussion environments to solve the simultaneous localization and mapping (SLAM) problem. But the standard PF suffers a lot from the sample impoverishment after resampling. This paper introduces a particle splitting technique before the resampling process, called pre-resampling. This method splits particles with big importance weight into several particles with small importance weight. The original particle set is regenerated, then we resample from the new particle set. The number of the particles will stays the same after resampling. The results of simulations show that the improved method mitigates sample impoverishment effectively comparing to the standard PF.